一种改进闭环检测算法的鼠类SLAM模型研究

A Rat SLAM Model Base on Improved Closed-loop Detection Algorithm

由于光线角度变化的干扰传统的鼠类SLAM模型导航适应性会变差,提出在鼠类SLAM模型基础上引入实时关键帧匹配的闭环检测算法,该算法通过存储同一定位点的不同签名更好的估计未来发生的闭环假设,提高了光线角度变化下场景的匹配率;同时,改进的闭环检测算法借鉴了研究发现人脑四种记忆模式的算法原理,选取一定数量具有良好特征的定位点实现闭环检测,提高了系统的实时性,降低了传统闭环检测算法的复杂度。将该算法融入现有鼠类SLAM模型中,分别从局部场景细胞学习的视觉模板、经验节点的匹配效果、以及绘制的经历图方面进行分析。实验结果表明融合实时关键帧匹配的改进闭环检测算法能够对光线角度变化的同一场景具有更强的鲁棒性,相比于传统闭环检测算法,系统实时性更佳。

The performance of the Rat SLAM model gets worse under the condition of the changing light, a closed-loop detection algorithm based on real-time key frame matching is proposed, this algorithm better estimates the closed loop assumption in the future by storing different signatures under the same location,which improves the matching rate of complex scenes under the circumstance of the angle of light changes. In the meantime, the improved closed-loop detection algorithm improves the real-time performance of the traditional closed-loop detection algorithm by referring to the mechanism of human brain memory. This algorithm is fused into the Rat SLAM model and experiments are done respectively from the visual template of local view cells, the matching effect of the experience nodes, and the experience map by the qualitative approach. Experiments show that compared with the traditional closed-loop detection, the improved closed-loop detection algorithm has stronger robustness under the circumstance of the angle of light changes and has better real-time performance.